Influence Score and Transformers interpretability: Measure of the Effective Impact of Attention Heads at inference time

Abstract

We propose an influence score to quantify the contribution of attention heads to classification decisions in Transformer-based models designed for prompt injection detection. The score combines directional influence on the logits with structural contribution within the residual stream, enabling a multi-scale analysis at the head, layer, and network levels. Applied to a DeBERTa model specialized for prompt injection detection, our framework reveals distinct decision behaviours between correct and

Categories

Framework mappings

Suggested from the entry's categories.

@misc{bouger2026influence,
  title = {{Influence Score and Transformers interpretability: Measure of the Effective Impact of Attention Heads at inference time}},
  author = {Lisa Bouger and Yannick Teglia and Philippe Loubet Moundi},
  year = {2026},
  month = sep,
  eprint = {2609.05074},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.05074}
}